Automatically detecting bregma and lambda points in rodent skull anatomy images.

Area of Science:

• Neuroscience and Biomedical Engineering
• Medical Imaging and Computer Vision

Background:

• Stereotactic neurosurgery relies on precise localization of injection sites for probes, cannulas, and optic fibers.
• Manual localization methods are prone to human error and variability, impacting experimental repeatability and treatment outcomes.

Purpose of the Study:

• To develop and validate an automated method for accurate localization of anatomical landmarks on rodent skulls.
• To reduce localization errors and enhance the repeatability of neurosurgical procedures in research.

Main Methods:

• Integration of a region-based convolutional network (R-CNN) and a fully convolutional network (FCN) for anatomical point identification.
• Utilizing rodent skull anatomy images to train and test the proposed localization framework.

Main Results:

• The automated framework successfully identified and located bregma and lambda in rodent skull images.
• Achieved mean localization errors below 300 μm, demonstrating high precision.
• The method exhibited robustness across varying lighting conditions and mouse orientations.

Conclusions:

• The proposed deep learning-based localization framework significantly reduces errors in identifying injection sites for stereotactic neurosurgery.
• This automated approach has the potential to streamline neurosurgical procedures and improve the reliability of experimental and therapeutic outcomes in rodent models.